🧠 Memory System

👤 Who it's for: All long-term users
⏱️ Read time: ~5 minutes
💡 In one line: Four memory types (core / user / daily / conversation) let AI remember preferences and context like a teammate.

YingClaw's persistent memory system gives AI continuity across sessions — the more you use it, the better it understands your preferences, habits, and work context.

Four Memory Types

Type	Lifespan	Purpose	Injection
core	Permanent	Core facts, project decisions, coding conventions	Auto-injected every session
user	Permanent	User preferences, personal info, communication style	Auto-injected every session
daily	7 days	Session notes, milestone progress	Queried on demand
conversation	Auto-managed	Automatically saved by the conversation system	System-managed

core — Permanent Core Memory

For storing infrequently changing key information:

• Project tech stack and architectural decisions
• Team coding standards and conventions
• Environment configuration essentials
• Workflow and process conventions

Keep each core memory entry under 100 characters for conciseness.

user — Permanent User Memory

Records personal user preferences:

• Programming language preferences ("User prefers Rust over Go")
• Communication style ("User likes concise replies, no verbose explanations")
• Common tools and work habits

daily — Short-Term Session Memory

For milestone information:

• Today's task progress
• Temporary debugging notes
• Short-term project status

Daily memories auto-expire after 7 days — no manual cleanup needed.

conversation — System-Managed

Conversation history automatically saved by the system. Do not manually manipulate this type.

Core Tools

memory_store — Save Memory

memory_store key="user_language" content="User prefers Rust and TypeScript" category="user"

Parameter	Type	Description
key	string	Unique identifier, use semantic naming
content	string	Memory content, keep it concise
category	string	Type: core / user / daily

memory_recall — Query Memory

memory_recall query="programming language preference" limit=5

Search saved memories by keyword, returns results sorted by relevance.

memory_forget — Delete Memory

memory_forget key="outdated_setting"

Delete a memory entry by key. Use when information is outdated or needs correction.

Memory Principles

Principle	Description
One fact per entry	Each memory stores a single fact — don't mix multiple pieces of info
Under 100 chars	Keep it concise for efficient injection and retrieval
User priority	User preferences and corrections take highest priority
No temporaries	Don't store temporary task progress or session state
Update promptly	Use memory_forget + memory_store to update stale entries immediately

Examples

Remember User Preferences

> "I prefer TypeScript over JavaScript"

→ memory_store key="lang_pref" content="User prefers TypeScript over JavaScript" category="user"

Save Project Environment Info

> "This project uses Rust 2024 edition, minimum MSRV is 1.75"

→ memory_store key="rust_project_config" content="Rust 2024 edition, MSRV 1.75" category="core"

Update After Correction

> "No, the API address I mentioned before is outdated, we use v2 now"

→ memory_forget key="api_base_url"
→ memory_store key="api_base_url" content="API endpoint updated to v2" category="core"

Notes

• Do not load private memories in group chat environments
• Credentials, keys, and other sensitive info should NOT be stored in memory
• Periodically review core and user memories, clean up outdated entries

Next Steps

Memory makes YingClaw understand you better. Next, explore 🛠️ Skill System to extend capability boundaries.